Abstract: In order to study the relationship between structure and fungicidal activity, in this article, based on 34 kinds of flavanone derivatives’ fungicidal activity which used as an indicator, stoichiometry software PCLIENT were used to obtain initial molecular descriptors for each compound, then, binary matrix shuffling filter (BMSF) and worst descriptor elimination multi-round (WDEM) method were successively applied to filter the selected descriptors, and 14 molecular descriptors were retained. Finally, we used multiple linear regression and support vector regression to build the model for all retained descriptors respectively, and analyzed the important influence of each retained descriptor based on the SVR nonlinear explanatory system. The model accuracy is higher based support vector nonlinear regression than multiple linear regression; the importance of 14 retained molecular descriptors is SssO > MATS6e > SeaC2C2aa > EEig04d > EEig03x > C-028 > R7v > L2e > X4A > Mor25m > G1m > Mor17m > BEHv1 > RDF045m, and obtains the high accuracy nonlinear SVR-QSAR model. The established model has certain reliability using our method, and plays the important role in conducting fungicides synthesis.